Spike: does the full e2e fit on GitHub's free runners? #7

Closed
opened 2026-11-18 05:14:37 +00:00 by jcoffey-dev · 3 comments
Owner

S7 · size S · part of inbuxa for Kubernetes (tracking issue in inbuxa-charts)

kind + CloudNativePG + NATS + Garage + server, webmail and console, then the mail test.

Done when

  • Full install and mail test in under 25 minutes on ubuntu-24.04 and ubuntu-24.04-arm
  • If not: what to cut in CI
**S7** · size S · part of inbuxa for Kubernetes (tracking issue in inbuxa-charts) kind + CloudNativePG + NATS + Garage + server, webmail and console, then the mail test. **Done when** - Full install and mail test in under 25 minutes on ubuntu-24.04 and ubuntu-24.04-arm - If not: what to cut in CI
jcoffey-dev added this to the 27.1 milestone 2026-11-18 05:14:37 +00:00
jcoffey-dev added the enhancement label 2026-11-18 05:14:37 +00:00
jcoffey-dev self-assigned this 2026-11-18 06:28:54 +00:00
jcoffey-dev added the
priority
p2
label 2026-11-18 06:28:54 +00:00
Author
Owner

Folded into the checklist on inbuxa/inbuxa-charts#14 to keep the issue list short. The details stay here; reopen this when work on it starts.

Folded into the checklist on inbuxa/inbuxa-charts#14 to keep the issue list short. The details stay here; reopen this when work on it starts.
Author
Owner

Partial answer (bench lab, 2026-11-18): it should fit; to confirm on GitHub once the repo is public.

  • lab/up.sh (kind with 1 control plane + 3 workers, CloudNativePG ×2, NATS ×3, Garage): 108 s on the bench, the first run including the node image pull.
  • First boot + registry setup + one server pod ready: about 1 minute more.
  • Memory with the server running: ~3.2 GB across the four kind nodes. GitHub's standard runners have 16 GB and 4 vCPUs.
  • Image loading needs docker save --platform linux/amd64 + kind load image-archive; plain kind load docker-image fails on multi-arch images.

Still to measure: the mail test (K-L2) and a run on ubuntu-24.04 / ubuntu-24.04-arm. GitHub's free minutes are for public repos, so this waits for rc.1 (or a one-off run on a private GitHub copy).

**Partial answer (bench lab, 2026-11-18): it should fit; to confirm on GitHub once the repo is public.** - `lab/up.sh` (kind with 1 control plane + 3 workers, CloudNativePG ×2, NATS ×3, Garage): **108 s** on the bench, the first run including the node image pull. - First boot + registry setup + one server pod ready: about 1 minute more. - Memory with the server running: **~3.2 GB** across the four kind nodes. GitHub's standard runners have 16 GB and 4 vCPUs. - Image loading needs `docker save --platform linux/amd64` + `kind load image-archive`; plain `kind load docker-image` fails on multi-arch images. Still to measure: the mail test (K-L2) and a run on `ubuntu-24.04` / `ubuntu-24.04-arm`. GitHub's free minutes are for public repos, so this waits for rc.1 (or a one-off run on a private GitHub copy).
Author
Owner

Answer: yes, with lots of room. The full e2e (lab/e2e.sh, inbuxa/inbuxa-k8s-lab#31) on GitHub's free runners, public repo, run 44679378072:

ubuntu-24.04 (amd64) ubuntu-24.04-arm
kind (1 + 3 nodes, Kubernetes 1.37), Kyverno, MetalLB, CloudNativePG ×2, NATS ×3, Garage 185 s 184 s
chart install: seed, server ×5, webmail ×2, console ×2 30 s 27 s
mail, webmail across replicas, console 2 s 2 s
e2e total 220 s 217 s
whole job, tools and checkout included 3 min 52 s 3 min 50 s
memory used at the end 3.9 GiB of 15 GiB 4.0 GiB of 15 GiB

The limit was 25 minutes, so nothing needs cutting in CI: two replicas of everything stays. Both architectures run the published multi-arch images (server rc.34, webmail rc.10, console rc.22). Public-repo runners are 4 vCPU / 16 GB; a private repo's would have been 2 vCPU, which is part of why the repos went public on 2026-11-20.

What it means for K-L8: a run is ~4 runner-minutes per architecture, so three Kubernetes minors × two architectures nightly is ~25 minutes a night, free on a public repo.

**Answer: yes, with lots of room.** The full e2e (`lab/e2e.sh`, inbuxa/inbuxa-k8s-lab#31) on GitHub's free runners, public repo, [run 44679378072](https://github.com/inbuxa/inbuxa-k8s-lab/actions/runs/44679378072): | | ubuntu-24.04 (amd64) | ubuntu-24.04-arm | |---|---|---| | kind (1 + 3 nodes, Kubernetes 1.37), Kyverno, MetalLB, CloudNativePG ×2, NATS ×3, Garage | 185 s | 184 s | | chart install: seed, server ×5, webmail ×2, console ×2 | 30 s | 27 s | | mail, webmail across replicas, console | 2 s | 2 s | | **e2e total** | **220 s** | **217 s** | | whole job, tools and checkout included | 3 min 52 s | 3 min 50 s | | memory used at the end | 3.9 GiB of 15 GiB | 4.0 GiB of 15 GiB | The limit was 25 minutes, so nothing needs cutting in CI: two replicas of everything stays. Both architectures run the published multi-arch images (server rc.34, webmail rc.10, console rc.22). Public-repo runners are 4 vCPU / 16 GB; a private repo's would have been 2 vCPU, which is part of why the repos went public on 2026-11-20. What it means for K-L8: a run is ~4 runner-minutes per architecture, so three Kubernetes minors × two architectures nightly is ~25 minutes a night, free on a public repo.
Sign in to join this conversation.